BrainHub

该数据集名为BrainHub，是一个多模态大脑理解基准，旨在分析大脑信号中所包含的信息，重点关注大脑字幕生成、定位、检索和视觉解码等任务。该数据集融合了来自NSD数据集和微软COCO的fMRI和图像对，用于各种任务，包括视觉概念识别和空间关系提取。规模上，数据集包含了四个被试者的数据，每人有24,980个训练样本。涉及的任务包括大脑字幕生成、大脑定位、大脑检索和视觉解码。

Named BrainHub, this dataset is a multimodal brain understanding benchmark designed to analyze the information contained in brain signals, with a primary focus on tasks such as brain caption generation, localization, retrieval, and visual decoding. It integrates paired fMRI data and images from the NSD dataset and Microsoft COCO for various applications including visual concept recognition and spatial relation extraction. In terms of scale, the dataset contains data from four human subjects, with each subject having 24,980 training samples. The tasks involved include brain caption generation, brain localization, brain retrieval, and visual decoding.